[LLVMbugs] [Bug 209] [loadvn/inline/scalarrepl] Slow optimizations with extremely large basic blocks

[bugzilla-daemon at cs.uiuc.edu]

http://llvm.cs.uiuc.edu/bugs/show_bug.cgi?id=209

sabre at nondot.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|REOPENED                    |RESOLVED
         Resolution|                            |FIXED
            Summary|[loadvn/inline/gcse] Slow   |[loadvn/inline/scalarrepl]
                   |optimizations with extremely|Slow optimizations with
                   |large basic blocks          |extremely large basic blocks



------- Additional Comments From sabre at nondot.org  2004-02-05 11:29 -------
Okay, with this final patch:
http://mail.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20040202/011170.html

The optimizer is finally able to handle the testcase respectably, taking ~2s to
optimize it (as opposed to ~300s originally).  I now get the following
-time-passes output:

  Total Execution Time: 2.0800 seconds (2.0652 wall clock)

   ---User Time---   --System Time--   --- Name ---
   0.8200 ( 39.6%)   0.0099 (100.0%)   Function Integration/Inlining
   0.5100 ( 24.6%)   0.0000 (  0.0%)   Global Common Subexpression Elimination
   0.4699 ( 22.7%)   0.0000 (  0.0%)   Scalar Replacement of Aggregates
   0.0599 (  2.8%)   0.0000 (  0.0%)   Bytecode Writer
   0.0500 (  2.4%)   0.0000 (  0.0%)   Combine redundant instructions
   ...

For reference, 'g++ -O2' takes 91s on the same testcase (with GCC 3.3).

-Chris



------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.